• Gaussian blur sigma=2, convert to frequencies before pixel FlowSOM
  • Pixel hClusters found using z-score cap=3, k=20 (no iterations, too many clusters if using thresh=2)
  • Cell-based clustering using frequencing of pixel FlowSOM clusters and 20 hClusters (cap): normalize by size, k=20
  • Only 88 pixel FlowSOM clusters

1 Histogram of counts in pixel clusters

2 Heatmaps of pixel clusters x markers

3 Compare pixel clusering with cell clustering

4 Cell-based FlowSOM clustering using pixel clusters

4.1 Cell cluster x frequency of pixel clusters

4.2 Cell cluster x marker (not weighted, segmentation)

4.3 Cell cluster x marker (weighted by pixel cluster frequency)

5 Cell-based FlowSOM clustering using hClusters (cap, k=20)

5.1 Cell cluster x frequency of pixel clusters

5.2 Cell cluster x marker (not weighted, segmentation)

5.3 Cell cluster x marker (weighted by pixel cluster frequency)

6 Heatmaps from normal segmentation